Nvidia Unveils Consumer RTX Spark PC Platform with N1X SoC, Shipping in October

Nvidia will begin shipping its RTX Spark devices in October 2026, powered by the N1X chip, which is purpose-built for local AI computing. The N1X supports up to 128GB of unified memory and integrates a Blackwell GPU capable of delivering up to a petaflop of floating-point performance, designed to meet AI inference demands on personal devices.

The N1X is offered in two configurations. The flagship version features a Blackwell GPU with 6,144 CUDA cores and a 20-core Grace CPU, supporting 24GB to 128GB of unified memory, targeting ultra-premium laptops and compact desktop systems. The second configuration reduces the GPU to 5,120 CUDA cores and the CPU to 18 cores, with unified memory ranging from 24GB to 32GB, aimed at traditional laptop form factors. The chip is based on the Arm architecture and tightly integrated with the Blackwell GPU.

Initial devices featuring the N1X are scheduled to launch in October 2026 through OEM partners including Acer, Asus, Dell, Gigabyte, HP, Lenovo, MSI, and Microsoft. The product lineup covers thin-and-light laptops and mini desktop PCs, with demo units from partner brands showing chassis thickness as low as 14 millimeters and claiming all-day battery life. Final pricing will be announced by each brand at launch.

Nvidia also introduced NVIDIA PAIR, an open-source personal AI router software. PAIR automatically discovers compatible devices on the same network and pools their compute resources to handle local AI inference tasks collaboratively. Compatible devices include PCs with RTX 20-series or newer graphics cards, RTX Pro GPUs, DGX Spark systems, and devices with Apple M4 or newer chips. PAIR prioritizes idle compute capacity and dynamically adjusts resource allocation based on device usage, such as reducing a device's compute contribution when it detects the user starting a game.

In an official demonstration, running an agent task using the Qwen 3.6 35B A3B model took 18 minutes on a single RTX Spark laptop. A three-device PAIR cluster consisting of one RTX Spark laptop, one DGX Spark, and one RTX 5090 device reduced the average time to 8 minutes and 48 seconds. Additionally, three AI agent applications—Perplexity Portable Computer, Hermes Agent, and OpenClaw—now offer simplified local deployment, allowing users to install and launch them with just a few clicks.